the terms of trade and economic growth in ......the terms of trade and economic growth in the...
TRANSCRIPT
THE TERMS OF TRADE AND ECONOMICGROWTH IN THE PERIPHERY 1870-1983
Christopher BlattmanJason Hwang
Jeffrey G. Williamson
Working Paper 9940
NBER WORKING PAPER SERIES
THE TERMS OF TRADE AND ECONOMICGROWTH IN THE PERIPHERY 1870-1983
Christopher BlattmanJason Hwang
Jeffrey G. Williamson
Working Paper 9940http://www.nber.org/papers/w9940
NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue
Cambridge, MA 02138August 2003
We acknowledge the superb research assistance of Martin Kanz. We are also greatly indebted to LuisBerrtola, Michael Clemens, John Coatsworth, and Yael Hadass for their help in constructing the underlyingdata base for other projects with Williamson. We are also grateful to them and to Michael Jansson, ChadJones, Ian McLean, Kevin O.Rourke, Leandro Prados, Martha Olney and members of the HarvardDevelopment and International Seminars for comments. Blattman and Hwang thank the Harvard Centerfor International Development and the National Bureau of Economic Research for office space andcomputing resources. Williamson thanks the National Science Foundation for support under NSF grantSES-0001362. The views expressed herein are those of the authors and not necessarily those of the NationalBureau of Economic Research.
©2003 by Christopher Blattman, Jason Hwang, and Jeffrey G. Williamson. All rights reserved. Short sectionsof text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit,including © notice, is given to the source.
The Terms of Trade and Economic Growth in the Periphery 1870-1938Christopher Blattman, Jason Hwang, and Jeffrey G. WilliamsonNBER Working Paper No. 9940August 2003JEL No. F1, N10, O1
ABSTRACT
The contending fundamental determinants of growth -- institutions, geography and
culture --exhibit far more persistence than do the growth rates they are supposed to explain.
So, what exogenous shocks might account for the variance around those persistent
fundamentals? The terms of trade seems to be one good place to look. Using a panel data base for
35 countries, this paper estimates the impact of terms of trade volatility and secular change between
1870 and 1938. We find that volatility was much more important than secular change. Additionally,
both effects were asymmetric between core and periphery, findings that speak directly to the terms
of trade debates that have raged since Prebisch and Singer wrote more than 50 years ago.
Christopher Blattman Department of Economics University of California Berkeley, CA [email protected]
Jason Hwang Department of Economics Harvard University Cambridge MA [email protected]
Jeffrey G. Williamson Department of Economics Harvard UniversityCambridge, MA 02138and [email protected]
1
1. Introduction
What role did secular terms of trade changes and its volatility play in explaining the level
and variance of country growth rates between 1870 and 1938? A decade ago, William Easterly,
Michael Kremer, Lant Pritchett and Larry Summers (1993) offered an insightful observation: all
the contending fundamental determinants of growth -- institutions, geography and culture --
exhibit far more persistence over time than do the growth rates they are supposed to explain. So,
where do we search for exogenous shocks that might account for the variance around those
persistent fundamentals? Exogenous relative price shocks associated with the external terms of
trade seem to be one good place to look, especially during world episodes of global integration
(or disintegration) when commodity prices converge (or diverge) world wide, inducing large
terms of trade changes and economy-wide responses. Most of the modern empirical analysis on
the growth and terms of trade connection has exploited recent evidence, where world commodity
price convergence has been more modest (Findlay and O�Rourke 2003). In contrast, the classic
Prebisch-Singer debate focused on the century before 1950, a period that included the first global
century up to 1913 and the autarky that followed, seven decades of dramatic commodity relative
price change.
It is also true that specialization in the production and export of primary products has
proven to be one of the most enduring and robust determinants of poor economic growth (Sala-i-
Martin 1997; Sachs and Warner 2001). However, the reason for this �resource curse� relationship
is poorly understood. Institutional or geographic features might foster natural resource
dependence. Or we could have it the other way around: natural resource activities might fail to
encourage the right sort of linkages, institutions or incentives for economic development (Tornell
and Velasco 1992; Engerman and Sokoloff 1997; Acemoglu, Johnson and Robinson 2003).
Perhaps, but could it also be that the relationship between natural resource dependence and
growth is somehow related to the terms of trade? Could the prices of primary products be more
2
volatile or trend differently than manufactures, and could this behavior somehow be tied to
growth performance?
This paper will use country experience both in the core and in the periphery and over the
seven decades before World War II to investigate whether variations in macroeconomic
performance can be attributed to secular change and volatility in the terms of trade. Following
Easterly et al.(1993), we will focus on just one aspect of macroeconomic performance, GDP per
capita growth rates. This paper is not the first to explore this relationship. In 1950 Raoul Prebisch
and Hans Singer argued that developing countries in the periphery (which exported primary
products and imported manufactures) had experienced both declining terms of trade and
stagnating incomes, and that the two phenomena were causally related. More recently,
economists have begun to argue that both terms of trade trends and volatility are related to
economic growth. The economics seems simple enough: a secular improvement in the terms of
trade leads to higher levels of investment, and hence long-run economic growth, while higher
volatility in the terms of trade reduces investment, and hence growth, because of aversion to risk
(Mendoza 1995, 1997; Deaton and Miller 1996; Kose and Reizman 2001; Bleaney and Greenway
2001).
In addition, one might well expect to find asymmetry between core and periphery. If the
secular increase in the terms of trade reinforces comparative advantage, then it induces more
industrialization in the core and less industrialization in the periphery. If industrialization is the
central carrier of growth, the secular terms of trade improvement should raise (long run) growth
rates in the core but lower them in the periphery. Similarly, rich countries with more sophisticated
institutions and markets are likely to have cheaper ways to insure against price volatility than
poor countries, so terms of trade instability is likely to have a far bigger negative impact in the
periphery than the core.
3
This paper examines these hypotheses relating the terms of trade and growth using data
for 35 countries over the years 1870 to 1938.1 The 35-country sample covers about 85 percent of
the world population in 1900, and it includes five industrial leaders (the USA, France, Germany,
Sweden and the UK), three rich European offshoots (Canada, Australia and New Zealand), nine
European industrial late comers (Austria-Hungary, Denmark, Greece, Italy, Norway, Portugal,
Serbia, Spain and Russia), eight primary product exporters in Latin America (Argentina, Brazil,
Colombia, Chile, Cuba, Mexico, Peru and Uruguay), and ten primary product exporters in Asia
and the Middle East (Burma, Ceylon, China, Egypt, India, Indonesia, Japan, the Philippines, Siam
and Turkey).
Using new terms of trade evidence, Figure 1 plots terms of trade trends across the five
regions. Note first the substantial variation across regions, and that a clear distinction between
core and periphery stands out: the core industrial leaders had rising terms of trade throughout the
seven decades; the periphery had no rise, and, indeed, Latin America and Asia both underwent a
long run decline.2 However, none of the regional secular trends appear to look very dramatic. As
we will see later in this paper, time-series analysis of country terms of trade provides very weak
support for a declining terms of trade in the periphery over the full seven decades. Regardless of
trend, however, we will see below that secular changes in the terms of trade had an impact on
growth in the periphery, but not in the core.
Figure 2 compares terms of trade volatility across these 35 countries in three sub-periods,
measured by the standard deviation.3 Generally, the industrial countries had lower terms of trade
volatility than did those in the periphery: only Germany and Italy are over towards the right, and
all of that is due to terms of trade volatility late in the period. Generally, the Asian and Latin
American primary product exporters had higher terms of trade volatility than did the European
1 The war years 1914-1918 are excluded due to a paucity of data and to the gross distortions to world trade and pricesattributable to war demands, blockades and skyrocketing transport costs.2 It was not a zero-sum game, since everybody�s terms of trade could have risen given that most of the period beforeWorld War I witnessed a dramatic fall in transport costs (Shah Mohammed and Williamson 2003).
4
core and its offshoots: only India, Burma and Peru are over towards the left. There are standard
explanations for that observed asymmetry in volatility, and they apply to our period as well.
Figure 3 looks at the volatility of different primary products. Clearly, not all commodities
were created with equal stability since some exhibit far greater volatility than others.4 To make
matters worse for the periphery, primary producers tend to have much higher commodity
concentration ratios, a second factor that accounts for the differences in volatility in country terms
of trade between core and periphery, and even within the periphery. Table 1 reports one measure
of export concentration for twenty-eight primary product exporting countries. The top two export
products are listed, along with the percentage of total exports they represent. While the level of
export concentration by this measure was, on average, about 80 percent, the range of export
concentration is quite wide. Countries like Argentina, Canada, India and Turkey had relatively
low levels of concentration, while Egypt, Ceylon, Chile, Columbia and Siam focused on just one
or two products throughout the seven decades. Countries with high export concentration usually
stayed that way, so that even when a new export was added to a country�s portfolio, typically an
old one dropped out.
But is higher volatility associated with lower growth? Figure 4 plots average annual GDP
growth rates and terms of trade volatility between 1870 and 1939, and visual inspection suggests
considerable support for the hypothesis that terms of trade instability lowers growth performance,
even when the relationship is unconditioned by other forces. Yet, not all countries in the
periphery were just primary product exporters. Many, as time went by, began to export textiles
and fuels.5 This can be seen in Table 2, which documents the level of primary product export
3 Throughout this paper we measure terms of trade volatility by the standard deviation. There is no need to divide bymean values since all the terms of trade series are indexed on the same base year.4 Carlos Diaz-Alejandro (1984) called this the �commodity lottery� when reviewing Latin American experience in the1930s.5 An even more dramatic retreat from export concentration and high primary product export shares has taken place inthe Third World since 1970. The World Bank reports for all �developing� countries that manufactures rose from only17.4 percent of commodity exports in 1970 to 64.3 percent by 1994. Enough of the Third World is now labor-abundantand natural-resource-scarce so that the growth of trade has helped it industrialize. It appears that the classic image ofThird World specialization in primary products is obsolescing (Lindert and Williamson 2003: p. 249).
5
shares in total exports (or primary product concentration) for the countries in the periphery.
Variance in primary product concentration may help account for the lack of an even stronger
unconditional correlation between volatility and growth in Figure 4. In addition, the size of the
export sector must have mattered. Given the same terms of trade volatility, countries with higher
X/GDP ratios would have suffered bigger falls in GDP per capita growth, if our hypothesis is
successful. Essentially, multivariate empirical analysis is needed to explore these relationships.
Therefore, this paper looks at cross-country growth regressions with secular terms of
trade changes and volatility as explanatory variables, and with export shares (lagged) and primary
product concentration (lagged) as controls. What we find confirms our hypotheses: terms of trade
volatility had a negative impact on growth in the periphery, but not in the core; and secular terms
of trade growth had a positive impact on growth, but it was much weaker in the periphery than in
the core. Does the strong explanatory power of terms of trade behavior on growth survive the
inclusion of primary product export concentration and export shares in GDP as control variables?
Indeed, they do. Something fundamental about natural resources links them to slow growth. The
resource curse stands.
The next section reviews the alternative theories linking the terms of trade to economic
development. Section 3 offers our empirical specification and the data. Section 4 presents the
results, and section 5 concludes.
2. Theory
2.1 Terms of Trade Volatility and Economic Growth
In the introduction to his famous 1950 article, Hans Singer proposed that fluctuations in the
terms of trade dramatically affected the funds available to underdeveloped countries for capital
formation, and hence growth. He noted that changes in the volume and value of foreign trade tend
to be important in underdeveloped countries because their surplus income over subsistence is
6
often entirely dependent on export revenues, and investment is in turn dependent on these income
sources. Unfortunately for Singer, he missed an opportunity by failing to dwell further on this
relationship between volatility and growth, and focused instead on secular trends in the terms of
trade.
Over the last decade, economists have begun to combine business cycle theory, growth theory
and data on macroeconomic volatility (in particular, trade shocks) to establish a relationship
between volatility and growth such as that discussed by Singer. In general, they have found that,
as Singer seemed to suggest, short-term volatility appears to be negatively correlated with
growth, perhaps by reducing levels of investment. Garey and Valerie Ramey (1995) review the
literature and examine the cross-country evidence between macroeconomic volatility and growth
using data from 92 developing and developed economies between 1962 and 1985. The authors
find that government spending fluctuations and volatility are significantly related, and that
countries with higher macroeconomic volatility have lower mean growth. Several recent papers
have investigated the specific relationship between terms of trade volatility and development,
relying primarily on African experience. Ayhan Kose and Raymond Riezman (2001) examine the
role of fluctuations in import and export prices in explaining macroeconomic fluctuations in 22
non-oil exporting African countries, 1970-1990. Constructing a multi-sector model of a small
open economy, and fitting African data to this model, they find that fluctuations in the prices of
these tradables account for roughly half of the fluctuations in aggregate output. Moreover, they
find that adverse shocks induce a significant decrease in aggregate investment. Angus Deaton and
Ronald Miller (1996) employ a different model but also conclude that terms of trade shocks play
an important role in macroeconomic fluctuations in African economies.
Enriquez Mendoza (1997) also proposes a stochastic growth model whereby terms of trade
uncertainty can adversely affect savings and growth. He develops a model where planned
consumption growth is an increasing function of terms of trade growth (because of the impact on
lifetime income), but a decreasing function of terms of trade volatility (because of risk
7
preferences). Cross-country panel regressions of 40 developed and developing countries seem to
support his model�s key predictions, and terms of trade shocks seem to account for nearly half of
actual GDP variability. Michael Bleaney and David Greenway (2001) obtain similar findings
with a panel of 14 sub-Saharan African countries. Both growth and investment increase when the
terms of trade improve, and both are negatively affected by terms of trade instability.
These studies point to a potentially robust negative relationship between economic growth
and terms of trade volatility. The channel favored by the literature seems to be through
investment and uncertainty. Moreover, the results carry over to a sample of developing countries
where investment often comes from abroad, not domestically. All of the above papers, however,
cover no more than two or three decades and focus on the last third of the twentieth century. It
seems to us instructive to examine these results over a longer period, and during an era when
primary product exports dominated the periphery even more than they do today.
2.2 Terms of Trade Trends and Economic Growth
The literature on the relationship between economic growth and secular trends in the
terms of trade is older and more contentious than that on the impact of terms of trade volatility. In
fact, there is no really well-articulated theory, let alone consensus, on the growth effects of long-
term trends in the terms of trade.
One set of voices predict a positive correlation between terms of trade and income
growth. Just above we noted Singer�s observation that increases in the terms of trade can provide
surpluses for long-term capital accumulation. The empirical studies reviewed seem to provide
support for this position (Bleaney and Greenway 2001; Kose and Reizman 2001; Mendoza 1997;
Deaton and Miller 1996). This claim, however, is not the one for which Singer�s paper is most
famous.
8
In 1950, along with Raoul Prebisch, Singer emphasized instead that the fundamental
nature of primary products and manufactures would, in the long run, cause primary product prices
to fall relative to those of manufactures. Primary product specializing countries would therefore
see deterioration in their terms of trade and, as producers of increasingly cheaper primary
products and consumers of increasingly expensive manufactures, a relative fall in incomes. The
Prebisch-Singer hypothesis obviously hinges on the premise that primary product commodity
prices will tend to fall relative to those of manufacturing over time. Prebisch and Singer argued
that manufacturing sectors generated monopoly profits, and that these profits would eventually
translate into real wage increases. Commodity markets, on the other hand, would see productivity
gains translated into a decrease in prices, rather than a rise in wages. Why? Because most
commodity markets are perfectly competitive, and demand for commodities is income-inelastic.
If elastic labor supplies were present in underdeveloped countries, they would only exacerbate the
problem (Lewis 1978: pp. 14-20).
The influence of structural differences between manufacturing and primary production on
the terms of trade has received some recent theoretical and empirical support. John Spraos (1983)
develops a theoretical model where, for a group of price and income inelastic commodities,
developing countries can be trapped by the specialization they inherited from previous
generations. In Prebisch-Singer style, productivity gains are translated into price decreases rather
than wage increases. More recently, Harry Bloch and David Sapsford (2000) develop a model of
price determination in primary product and manufacturing sectors, where wages and prices in
primary production are treated as competitively determined while those in manufacturing are
determined by mark-up pricing and union-employer bargaining. Using price and wage data from
the post-World War II period they find support for the Prebisch-Singer hypothesis of declining
prices of primary products relative to those of manufactures.
Other theories also predict a negative correlation between the terms of trade and growth,
but for different reasons. Jeffrey Sachs and Andrew Warner (1995, 2001), for example, note that
9
countries with great natural wealth tend to grow more slowly than resource-poor countries:
natural resources, and any terms of trade boom that raises their value, are therefore a �curse� to
development. In cross-country regressions, specialization in primary products has proven to be
one of the most robust determinants of slow economic growth (Sala-i-Martin 1997). No one
theory of the natural resource curse is universally accepted, however. Sachs and Warner use the
more popular �crowding-out� logic, whereby primary production crowds out growth-enhancing
alternative activities. They suggest that manufactures production is the crowded out activity in
many resource-rich economies. A political economy approach offers an alternative to the
crowding-out theory. Thus, Anne Krueger (1974) famously argued that rent-seeking was a
growth-suppressing tendency of resource-owning elites in poor countries. More concretely, Aaron
Tornell and Andres Velasco (1992) suggest that resource rich poor countries have undeveloped
property rights, so that gains (in particular rents) are transferred to rich countries for safekeeping.
Hence terms of trade booms translate into capital flight.
There is some evidence to support the proposed negative correlation between secular
terms of trade improvements and economic growth in the primary-product-producing periphery.
Yael Hadass and Jeffrey Williamson (2003), for example, find that while terms of trade
movements between 1870 and World War I favored primary product exporters, it reduced their
growth. They also find strong evidence of asymmetry in growth impact between core and
periphery. Their sample, however, covers few of the developing countries that remained poor up
to World War II, and they did not explore the influence of volatility. A large sample of
underdeveloped countries has yet to be used to test the influence of terms of trade growth and
volatility during the period that motivated the Prebisch-Singer debate in the first place.
10
3. Empirical Specification and Data
Following the research strategy of Mendoza (1997), we examine the relationship between
economic growth and changes in the level and the volatility of the terms of trade in ten-year
periods over a span of seven decades, 1870-1938.
3.1 Empirical Strategy
Specifically, we regress 10-year average GDP per capita growth rates on 10-year average
terms of trade growth and the 10-year standard deviation of terms of trade, controlling for initial
income per capita, export shares, primary product export shares in total exports and a variety of
other variables. The basic regression model is as follows:
GRyit = β0 + β1 ln y it + β2 GRTOT it + β3 SDTOT it + β4 X it + β5 Yt + β6 Zi + εit (1)
for country i and decade t. GRy is the 10-year average growth rate in GDP per capita in
percentage units, GRTOT represents the 10-year average growth in the terms of trade (also in
percentage units), SDTOT is the standard deviation of the terms of trade over the 10-year interval,
and ln y is the natural log of GDP per capita in the first year of the decade. Other control
variables, represented above by the X vector, include the share of primary product exports in total
exports, which we will refer to by PP/X, and the export share in GDP, denoted X/GDP. We will
also interact variables in the X vector with GRTOT and SDTOT. Export concentration is not
included here since its alleged impact is already embedded in the measured SDTOT. To test for
asymmetry, we will break the sample up into center and periphery, and within each include the
interaction with the share of primary product exports in total exports. To control for multiplier
effects, we will also interact terms of trade growth and volatility with initial export shares in
GDP. The data for the pre-World War II period is not sufficiently comprehensive to add most of
the explanatory variables typical of the new empirical growth literature. However, we do not see
this as a shortcoming since our purpose is to account for the deviation of growth rates around
11
some steady state, and while understanding fundamentals driving that steady state is not the
motivation here, we do impose country fixed effects to capture all of those fundamentals. In any
case, we will also control for primary school enrollment rate. We experimented with the growth
in the land-labor ratio (as a measure of changing endowments, suggested by Hadass and
Williamson 2003), and with various measures of openness (tariff rates, local railway networks,
and transport costs from port to port), but since their inclusion had no effect on our variables of
interest they are omitted from the reported regressions. Finally, and as mentioned above, we
include time and country fixed effects (represented by Y and Z above) in order to control for
unobserved characteristics in countries and decades.
3.2 Data
Appendix 1 lists the sources and method of construction of all the variables used in this
paper. Still, given the centrality of the growth and the terms of trade data to our investigation, it is
worth discussing here the principal sources and techniques used in their construction. In addition,
we start by defining membership in core and periphery.
3.2.1 Defining Core and Periphery
Our 35 countries must be allocated between core and periphery. The allocation is
achieved by applying three criteria: the share of exports primary products; level of development;
and achieving similar sample sizes. The allocation has resulted in the following: Core = 19 = 4
industrial leaders (France, Germany, UK, USA), 5 rich European �frontier� offshoots (Argentina,
Australia, Canada, New Zealand, Uruguay), and 10 European industrial late comers (Austria-
Hungary, Denmark, Greece, Italy, Norway, Portugal, Serbia, Spain, Sweden, Russia); Periphery
= 16 = 6 primary product exporters in Latin America (Brazil, Colombia, Chile, Cuba, Mexico,
Peru), and 10 primary product exporters in Asia and the Middle East (Burma, Ceylon, China,
Egypt, India, Indonesia, Japan, the Philippines, Siam, Turkey).
12
3.2.2 GDP Per Capita Data
For most countries in our sample, the primary source of the GDP per capita estimates (in
1990 $US) is Angus Maddison (1995). Since Maddison�s developing country GDP estimates
often begin only with 1900 or even 1913, estimates for earlier years must be obtained from
supplementary sources, in particular backcasting from the Maddison 1900 or 1913 benchmarks
by using estimates of real wage growth constructed by Williamson. Full details are available in
Appendix 1. The doubtful quality of the GDP per capita growth estimates for some countries, and
especially for the late 19th century, implies that measured changes in GDP per capita over a
decade are almost certainly more reliable than annual estimates, a key consideration in electing to
use decadal rather than annual regressions. They will also give a better fix on long run effects.
3.2.3 The Terms of Trade Data
The net barter terms of trade is, of course, defined as the ratio of export to import prices.
It is not meant to be a welfare measure, but rather an index of the relative price shocks to which
the economy must adjust. Seven of our sampled European or European offshoot countries have
excellent terms of trade data, but this is not true of the rest. Thus, we constructed new terms of
trade series for the remaining twenty-eight, sixteen of which being commodity-exporting
countries in the periphery. Formally,
for product i, country j, and period t.
These TOT estimates differ from series generated by previous scholars in three key
∑
∑
=
=
⋅
⋅= J
jjtj
J
jjitji
ti
wp
wpTOT
1,
1,,,
,
13
respects.6 First, we employ country-specific export price indices in the numerator, not some
global commodity price index summarizing average price experience of manufactures or of
primary products. Second, we employ an international market price for both numerator and
denominator. To the extent that transport costs and tariff rates change significantly over time, a
country�s terms of trade measured in home markets might have obeyed somewhat different laws
of motion than ours having been measured in international markets. While our indices might not
exactly represent domestic prices, we are confident that the two sources would yield very similar
SDTOT as well as trends in TOT, especially after 1900 when the decline in ocean transport costs
slowed down considerably (Shah Mohammed and Williamson 2003). In any case, international
prices have the advantage of greater accuracy and availability. Third, the import index in the
denominator is in fact a price index for US manufactured goods, extensively traded.7 The same
import price index has been used for all periphery countries since reliable country-specific import
mix data are not available for most periphery countries before World War I. What data do exist,
however, suggest that the assumption closely approximates reality. Moreover, Kose and Riezman
(2001) suggest that the use of such a price index is superior to an import index when examining
the effects of trade shocks, and that such an index exhibits similar (albeit slightly higher) levels of
volatility compared with a pure terms of trade measure.
Given the extensive, and some would say exhaustive, effort that has gone into the
construction of existing US and European terms of trade series, we have opted to employ terms of
trade indices from traditional secondary sources for the industrial core and the European
offshoots. These traditional sources have not necessarily been constructed using the same price
sources and weighting schemes, and so are not strictly comparable to our indices constructed for
the periphery. Since the primary object of our investigation are the periphery countries -- the core
6 The literature dealing with the secular terms of trade facts is huge. However, most time series work over the past 15years has been based on the 1900-1986 Grilli and Yang (1988) series and the extension to 1991 by Bleaney andGreenway (1993). However, these series are not country based.7 The index is a weighted sum of the prices of textiles (55%), metals (15%), machinery (15%), building materials(7.5%), and chemicals and pharmaceuticals (7.5%) from the US Department of Commerce Historical Statistics.
14
countries will be considered for comparative proposes only, we do not think differences in terms
of trade construction pose a serious problem. Moreover, construction techniques seem broadly
similar and hence fairly comparable. These and other aspects of the terms of trade data are
described further in Appendix 1.
4. The Results
The impact of both the secular changes and the volatility of the terms of trade are
estimated in Table 3. Note that the results are reported for both the full seven decades (1870-
1938) and for that portion of the periphery�s terms of trade experience that was the prime focus of
Prebisch and Singer (1890-1938). In both cases, we omit the decade of World War I for the
reasons discussed earlier. Regarding hypothesis testing, the qualitative results do not differ at all
between the shorter and the longer period (although significance is somewhat better for the 1890-
1938 period since, one supposes, the data are better). Note also that the results are reported
separately for core and periphery, thus making it possible to test the asymmetry hypothesis. Table
3 is estimated using country and time fixed effects for the reasons offered earlier, namely it
makes it possible for us to control for the (unobserved) fundamentals that were also determining
growth performance, fundamentals that are not the focus of this paper. The unit of observation is
a decade, and thus the dependent variable is average annual growth over some decade for the
country in question. Similarly, % ToT refers to the growth (or decline) in the terms of trade over
that decade while sd ToT is the average volatility of the terms of trade over the decade. The other
independent variables (ln GDP per capita, PP/X, X/GDP, % Kids in School) are taken from the
first year of that decade, in an effort to avoid problems of causality and endogeneity.8 In the
periphery, we have data for every country and every decade, giving us 96 (=16x6) observations.
8 Since lagging X/GDP may not be enough to satisfy those who still worry about the endogenity of export shares, wereport in Appendix 2 the regressions in Table 3 without the export share variable interacting with %ToT and sd (ToT).The qualitative results of Table 3 are confirmed.
15
There are a few missing observations in the core, leaving us with 110 observations, as opposed to
114 (=19x6) that would be in a complete dataset.
Regarding controls, here are the key results. First, there was strong conditional
convergence in both regions, the coefficient on initial GDP per capita being negative, significant
and of considerable economic size. However, this result is for country fixed effects, so it refers to
convergence after controlling for the fundamentals. We know, of course, that those fundamentals
differed so dramatically across these countries that (unconditional) divergence took place �big
time� before 1938 (Pritchett 1997; Bourguignon and Morrisson 2002), so we do not find the
estimated conditional convergence very interesting. Second, enrollment rates did not have a
significant impact on growth performance. Indeed, the only time the enrollment rate is significant
(periphery, 1890-1938), it is of the wrong sign.9 Third, natural resources were not an
unambiguous �curse� by themselves: only for the periphery 1890-1938 was primary product
specialization (e.g. large PP/X) associated with poor growth. What made primary product
specialization a curse for the low-income economy in the periphery was how that specialization
influenced the terms of trade.
To identify terms of trade effects properly, we interact it with the other conditions that we
argued earlier should have mattered. Thus, while % ToT has by itself a positive impact on GDP
growth (but significantly so only for the periphery), and while it is especially big for those
countries with big trade shares (but significantly so only for the periphery), the impact of % ToT
is negative when interacted with PP/X. Thus, those countries specializing in primary product
exports received smaller beneficial effects from terms of trade gains than did those specializing in
manufactured exports (and especially so within the periphery). The critical question is whether
that offset was big enough to swamp all other positive terms of trade effects, and an answer to
9 There are two possible explanations for this apparently perverse result: the influence of schooling is already capturedby country fixed effects; and/or the schooling data are poor for the periphery before 1938.
16
that question will emerge when in Table 4 we add up the three coefficients on variables including
% ToT, using appropriate weights.
Note that the asymmetry hypothesis is borne out powerfully in Table 3. Indeed, nowhere
in Table 3 does the terms of trade have a significant effect on growth in the core, either for
secular change or for volatility. In contrast, for the periphery the terms of trade coefficient is
significant four out of five times for 1870-1938, and always significant for 1890-1938.
Finally, Table 3 suggests that terms of trade volatility had a far more powerful negative
impact on growth in the periphery than did secular trends. To see this more clearly, we need to
add up % ToT and sd ToT effects. Table 4 uses mean values of our explanatory variables and the
estimated coefficients from Table 3 to assess the marginal and actual impact of both % ToT and
sd(ToT). The marginal impact represents the impact of a one-unit increase in either secular
change or volatility on income growth. For % ToT, for example, it is the sum of the coefficient
estimates on % ToT, % ToT x PP/X and % ToT x X/GDP using as weights for the interacted
terms mean values of PP/X and X/GDP.10 The actual impact calculates the full impact of terms of
trade behavior by multiplying each coefficient estimate by the observed mean value of the
corresponding explanatory variable for the region and period in question. Thus differences in
marginal impact relate to country structural attributes that would generate different responses to
the same terms of trade movement, while differences in actual impact reflect both structural
elements as well as the actual terms of trade experience of each region.11
Looking at Table 4, an immediate finding is that the actual impact calculations strongly
support the asymmetry hypothesis. Over the full 1870-1938 period, secular changes in the terms
of trade served to diminish GDP per capita growth rates in the periphery (by 0.2 percentage
10 Marginal impact would obviously be just the coefficient estimate if our regression model included only the linearterms of % ToT and sd ToT without the interaction terms. Doing this for the periphery for 1870-1938, the coefficientestimates are 0.047 for % ToT and -0.66 for sd ToT with t-statistics of 0.51 and -1.66 respectively. Table 4 shows thatthe marginal impacts calculated from the model used in Table 3 for the same region and period are 0.153 and -0.046each with much greater statistical significance. We interpret these results to mean that there is an important role for theinteraction terms in understanding the impact of terms of trade movements.
17
points) but not in the core. Over the same period, terms of trade volatility also served to diminish
growth rates in the periphery (by 0.5 percentage points), but not in the core. Further inspection of
Table 4 shows that both differences in terms of trade movements and in marginal impact of terms
of trade are responsible for the regional asymmetry in actual impact. That is, the periphery stood
to gain from positive trends in terms of trade but the terms of trade deteriorated, producing a
negative total impact on growth. In contrast, terms of trade had little impact in the core, both
because of the small magnitude of the secular trends, and also because the underlying structure of
the economies made the marginal impact small. Similarly, the marginal impact of terms of trade
volatility was significantly negative in the periphery while it was zero in the core. The gap in
actual impact was, again, further widened by the fact that terms of trade was more volatile in the
periphery.
Second, the negative impact on growth from terms of trade volatility was two and a half
times greater than from secular trends (-0.501 versus �0.201). Further, the economic effects were
big. Terms of trade trends and volatility combined served to lower annual growth rates in the
periphery by more than 0.7 percentage points, a very big number when compared with the mean
growth rate in the periphery 1870-1938 of 0.8 percent per annum (Table 4, first panel). That is,
had there been no trend in the terms of trade and had the terms of trade been completely stable,
GDP per capita in the periphery would have grown at something like 1.5 percent per annum
between 1870 and 1938, a growth rate that would have more than matched that of the core (1.39
percent per annum). While this counterfactual does not imply any significant catching up of
periphery on the core, it does imply that there would not have been any �big time� divergence.
What if the periphery had experienced the same trend and volatility in terms of trade as the
industrialized core? The estimates suggest that the combined impact would have been to reduce
11 Impact calculations for the core are for illustrative purposes only, since the coefficient estimate are never statisticallysignificant.
18
the growth rate by 0.37 percent per annum,12 which is about half of the true actual impact. Thus if
the periphery had undergone the same terms of trade trends and volatility as the core, then terms
of trade experience would still have been a force for divergence in income levels, although with a
smaller influence than was actually the case.
5. Concluding Remarks
Consistent with recent studies of modern cross-country growth performance, our analysis
of a near century of pre-World War II data has shown that terms of trade movements were a very
important determinant of country economic performance. They were especially important in the
periphery where those less developed countries were, according to our empirical analysis, much
more sensitive to terms of trade trends and volatility than was true of industrial countries in the
core. The secular deterioration in the terms of trade experienced by the periphery represented a
significant drag on income growth during those seven decades 1870-1938. But even more
damaging to the primary product producers in the periphery was the high degree of volatility in
the terms of trade that exerted a negative impact on growth more than twice the size of the
negative impact of the trend. The two combined served to halve the growth performance of the
periphery.
What is especially notable about our findings is the presence of striking asymmetry
between the core and the periphery. While the terms of trade appears to have played an important
role in explaining (disappointing) growth performance in the less industrialized periphery, neither
trends nor volatility seems to have mattered much in the more industrial core, despite the fact that
volatility was almost as high in the core as in the periphery. Exactly what was it about the more
12 Because the regression model includes interaction terms, we need to make an assumption about the covariancebetween the interacted variables. We assume a zero covariance, so that, for example, we obtain a counterfactual valuefor % ToT x PP/X by simply multiplying the observed mean value of % ToT of the core and PP/X of the periphery. Webelieve this is reasonable because comparing the mean of the interaction term with the means of the interacted variablesseparately suggests that the covariance is small. Further, one of the independent variables, X/GDP, varies very littleacross the regions, implying that our assumption is likely to be innocuous.
19
industrialized countries that allowed them to escape the damaging consequences of terms of trade
instability, an escape that was apparently unavailable to primary product exporters, most of whom
were in the periphery? Did the industrialized economies have a better mechanism by which to
insure against adverse shocks? Were international capital markets important for smoothing
accumulation performance, and were these more accessible to core countries? Similarly, why was
it that countries in the core did not benefit much when the terms of rose, or suffer much when
they fell? Finally, while we have taken the terms of trade to be exogenous, future work might be
advised to explore the deeper question: What were the sources of the cross-country differences in
terms of trade trends and volatility? In particular, why did some countries experience large
swings in terms of trade while others enjoyed greater stability?
Our next step will be to assess the channel of impact that we have assumed here to be
accumulation. We have in hand a data set that documents British capital exports to all of the
countries in our sample up to 1913 (Clemens and Williamson 2003), after which global capital
markets fall apart. A future version of this paper will speak to this issue by reporting the influence
of terms of trade volatility on the late 19th century allocation of British capital across recipient
countries. In the meantime, we hope that this paper will persuade others to turn their attention to
the country impact of terms of trade shocks � rather than just the shocks themselves.
20
References
Acemoglu, D., S. Johnson and J.A. Robinson, (2003), �Institutions and Economic Development,�
unpublished manuscript.
Bleaney, M. and D. Greenway (1993), �Long-Run Trends in the Relative Price of Primary
Commodities and in the Terms of Trade of Developing Countries,� Oxford Economic Papers
45: 349-63.
Bleaney, M. and D. Greenway (2001), �The Impact of Terms of Trade and Real Exchange Rate
Volatility on Investment and Growth in Sub-Saharan Africa,� Journal of Development
Economics 65: 491-500.
Bloch, H. and D. Sapsford (2000), �Whither the Terms of Trade? An Elaboration of the Prebisch-
Singer Hypothesis,� Cambridge Journal of Economics 24(4): 461-81.
Bourguignon, F. and C. Morrisson (2002), �Inequality Among World Citizens 1820-1992,�
American Economic Review 92 (September): 727-44.
Clemens, M. A. and J. G. Williamson (2003), �Wealth Bias in the First Global Capital Market
Boom, 1870-1913,� Economic Journal (forthcoming).
Deaton, A. and R. I. Miller (1996), �International Commodity Prices, Macroeconomic
Performance and Politics in Sub-Saharan Africa,� Journal of African Economics 5: 99-191,
Supplement.
Diaz-Alejandro, C. (1984), �Latin America in the 1930s,� in R. Thorp (ed.), Latin America in the
1930s (Nlondon: Macmillan): 17-49.
Easterly, W., M. Kremer, L. Pritchett, and L. H. Summers (1993), �Good Policy or Good Luck?
Country Growth Performance and Temporary Shocks,� Journal of Monetary Economics 32:
459-83.
21
Engerman, S. and K. Sokoloff (1997), �Factor Endowments, Institutions, and Differential Paths
of Growth Among New World Economies,� in S. Haber (ed.) How Latin America Fell Behind
(Stanford, CA: Stanford University Press): 260-304.
Findlay, R. and K. H. O�Rourke (2003), �Commodity Market Integration, 1500-2000,� in M. D.,
Bordo, A. M. Taylor and J. G. Williamson (eds.), Globalization in Historical Perspective
(Chicago: University of Chicago Press): 13-62.
Grilli, E. R. and M. C. Yang (1988), �Primary Commodity Prices, Manufactured Goods Prices,
and the Terms of Trade of Developing Countries: What the Long Run Shows,� World Bank
Economic Review 2: 1-48.
Hadass, Y. and J. Williamson (2003) "Terms of Trade Shocks and Economic Performance 1870-
1940: Prebisch and Singer Revisited," Economic Development and Cultural Change 51(3):
629-56.
Kose, M. A. and R. Reizman (2001), �Trade Shocks and Macroeconomic Fluctuations in Africa,�
Journal of Development Economics 65(2): 55-80.
Krueger, A. O. (1974), �The Political Economy of the Rent-Seeking Society,� American
Economic Review 64 (June): 291-323.
Lewis, W. A. (1978), The Evolution of the International Economic Order (Princeton, NJ:
Princeton University Press).
Lindert, P. H. and J. G. Williamson (2003), �Does Globalization Make the World More
Unequal?� in M. D. Bordo, A. M. Taylor, and J. G. Williamson (eds.), Globalization in
Historical Perspective (Chicago: University of Chicago Press).
Maddison, A. (1995), Monitoring the World Economy 1820-1992 (Paris: OECD Development
Centre Studies)
Mendoza, E. (1997), �Terms of Trade Uncertainty and Economic Growth,� Journal of
Development Economics 54: 323-56.
22
Michaely, M. (1962), �Concentrations In International Trade,� Contributions to Economic
Analysis (Amsterdam: North Holland Publishing Company).
Prebisch, R. (1950), The Economic Development of Latin American and Its Principal Problems
(Lake Success, NY: United Nations, Department of Economic Affairs). Reprinted in
Economic Bulletin for Latin America 7 (1962): 1-22.
Pritchett, L. (1997), �Divergence, Big Time.� Journal of Economic Perspectives 11 (Summer): 3-
18.
Ramey, G. and V.A. Ramey (1995), �Cross-country Evidence on the Link Between Volatility and
Growth,� NBER Working Paper 4959, National Bureau of Economic Research, Cambridge,
Mass.
Sachs, J. and A. Warner (1995), "Economic Reform and the Process of Economic Integration,"
Brookings Paper on Economic Activity 1: 1-53
Sachs, J. and A. Warner (2001), �The Curse of Natural Resources,� European Economic Review
45 (May): 827-38.
Sala-i-Matin, X. (1997), �I Just Ran Two Million Regressions,� American Economic Review
87(2): 178-83.
Shah Mohammed, S. and J. G. Williamson (2003), �Freight Rates and Productivity Gains in
British Tramp Shipping 1869-1950,� NBER Working Paper 9531, National Bureau of
Economic Research, Cambridge, Mass. (February).
Singer, H. W. (1950), "The Distribution of Gains between Investing and Borrowing Countries,"
American Economic Review 40: 473-85.
Spraos, J. (1983) Inequalising Trade? (Oxford: Clarendon Press).
Tornell, A. and A. Velasco (1992), �The Tragedy of the Commons and Economic Growth: Why
Does Capital Flow from Poor to Rich Countries?� Journal of Political Economy 100 (1992):
1208-31.
23
Appendix 1: The Data
Most data used in this paper come from a novel database constructed by collaborations betweenJeffrey Williamson and Luis BJrtola, Chris Blattman, Michael Clemens and Yael Hadass, oftenfrom primary sources at the Harvard library. This data appendix attempts a thorough descriptionof sources and methods used in collecting these data, although much of the data herein haveappeared in previous publications and more detailed descriptions of sources and methods areavailable within these. Most data for the period 1870-1913 first appeared in Clemens andWilliamson, �Where did British Foreign Capital Go?� NBER Working Paper 8028, NationalBureau of Economic Research, Cambridge, Massachusetts (December 2000). Most data for theperiod 1914-1940 first appeared in Clemens and Williamson, �Why the Tariff-GrowthCorrelation Changed After 1950,� NBER Working Paper 8459, National Bureau of EconomicResearch, Cambridge, Massachusetts (October 2001). Data for the 8 Latin America countries,however, was updated and expanded in Coatsworth and Williamson (2002) "The Roots of LatinAmerican Protectionism: Looking Before the Great Depression," NBER Working Paper 8999,June 2002, which has a complete listing of changes and additions. Note that several sources areused frequently in all four aforementioned papers. They are: Arthur S. Banks, Cross-NationalTime Series,1815-1973, [Computer File] ICPSR ed. (Ann Arbor, Michigan: Inter-UniversityConsortium for Political and Social Research, 1976), hereafter Banks (1976); Brian R. Mitchell,International Historical Statistics, Europe, 1750-1988 (New York: Stockton Press, 1992); BrianR. Mitchell, International Historical Statistics: The Americas, 1750-1988 (New York: StocktonPress, 1993); Brian R. Mitchell, International Historical Statistics: Africa, Asia & Oceania,1750-1993 (New York: Stockton Press, 1998). hereafter Mitchell.
GDP and GDP per capitaThe units on this variable are 1990 US dollars per inhabitant of any age. For most countries,
gross domestic product per capita (in 1990 US dollars) is taken from Angus Maddison,Monitoring the World Economy 1820-1992 (Paris: Development Center of the Organization forEconomic Cooperation and Development, 1995).
GDP per capita estimates for 1870-1950 for Australia, Brazil, Canada, China, Denmark,France, Germany, India, Indonesia, Italy, Japan, Mexico, New Zealand, Norway, Portugal,Russia, Spain, Sweden, Siam, and the United States come from Angus Maddison, 1995,Monitoring the World Economy, 1820-1992, OECD, Paris. For countries not reported inMaddison (1995), GDP per capita is calculated by dividing a country�s income (in 1990 USdollars) by population in every year. Sources of the population data have been describedelsewhere in this appendix, and the sources of the income estimates follow.
Data for Argentina after 1890 come from Maddison op. cit. Before this date, GDP per capitais assumed to grow at the same year-on-year rate as the estimates of Argentine real wages foundin Jeffrey G. Williamson, 1995, �The Evolution of Global Labor Markets since 1830:Background Evidence and Hypotheses,� Explorations in Economic History, 32:141-196.
GDP per capita estimates for Austria-Hungary before 1914 come from David F. Good, 1994,�The Economic Lag of Central and Eastern Europe: Income Estimators for the HabsburgSuccessor States, 1870-1910,� Journal of Economic History, 54(4)(December): 69-891. Theseare converted from 1980 to 1990 dollars using a GDP deflator obtained from the Bureau ofEconomic Analysis of the United States Department of Commerce (online athttp://www.bea.doc.gov/bea/dn/gdplev.htm).
Data for Burma after 1900 come from Maddison op. cit. Before this date it is assumed thatBurmese growth mirrored that of India.
24
Ceylon presented the most difficult data challenge in this category, as we are not aware ofany published figures for GDP in Ceylon during this period. Campbell (Burnham O. Campbell,1993 �Development Trends: A Comparative Analysis of the Asian Experience,� in NaohiroOgawa et al., eds., Human Resources in Development along the Asia-Pacific Rim, OxfordUniversity Press, New York) has estimated that in 1914, GDP per capita in Ceylon was 1.95times that of India. The same ratio had declined to 1.52 by 1948 according to United NationsStatistical Yearbook 1949-50 (New York: 1950), pp. 21-2 and 406. In the intervening years,1914-1948, it is assumed that the ratio declined annually at a constant rate. Before 1914, it isassumed that real GDP per capita grew at the same rate as did the ratio of the real value of Britishcolonial revenue from Ceylon to the population of the Island. A full series of annual nominalcolonial revenues and population figures come from the 1905 and 1914 editions of the annualCeylon Blue Book, a statistical publication of the colonial administration in Colombo. Some ofthese figures were recorded in rupees, and are converted to pounds sterling using conversion ratesfrom Bryan Taylor II, 2000, Encyclopedia of Global Financial Markets, Global Financial Data,Los Angeles, California (online at http://www.globalfindata.com). The resulting figures areconverted to real pounds sterling using the deflator in McCusker op. cit.
Data for Chile after 1900 come from Maddison op. cit. Before this date it is assumed thatChile grew at the same year-on-year rate as did our estimates of Argentine GDP per capita.
Data for Colombia after 1900 come from Maddison op. cit. Before this date, it is assumedthat that GDP per capita grew at an unweighted average of the growth rates for Mexico and Brazilbetween 1850 and 1900 given in Coatsworth op. cit.
Estimates for Cuba for 1850 and 1913 are based on estimates of Cuban GDP per capitarelative to that of Mexico and Brazil presented in John H. Coatsworth, 1998, �Economic andInstitutional Trajectories in Nineteenth-Century Latin America,� in John H. Coatsworth and AlanM. Taylor, eds., Latin America and the World Economy Since 1800, Harvard University Press,Cambridge, Mass. An unweighted average of the figures implied by Coatsworth�s proportion ofour estimates for Mexico and Brazil is calculated for both years, and the intervening yearsestimated by geometric interpolation. For the years 1914-1950, Cuba�s Net National Product incurrent year pesos comes from Mitchell (1993). These NNP values are converted to 1990 USdollars with the help of the peso-dollar exchange rate given in Taylor (2000) and the Americanhistorical consumer price index given in John McCusker, How Much Is That in Real Money(Worcester, Mass.: American Antiquarian Society, 1992), pp. 330-2.
Estimates for Egypt after 1900 come from Maddison. Before this date it is assumed that GDPper capita grew at the same year-on-year rate as did estimates of Egyptian real wages from JeffreyWilliamson, 2000, �Real wages and relative factor prices around the Mediterranean, 1500-1940,�in Şevket Pamuk and Jeffrey G. Williamson, eds. The Mediterranean Response to GlobalizationBefore 1950, Routledge, New York. For the years 1900-1950, a trend for Egyptian GDP percapita is calculated with the help of benchmark values given in Maddison (1995). Annual GDPper capita estimates are then calculated under the assumption that Egypt deviated from theMaddison-estimated Egyptian benchmark trend in the same way (percentage-wise) as Turkey didfrom her GDP per capita trend (after the civil war).
Data for Greece are estimated by projecting Maddison�s (op. cit.) 1913 figure backwards,assuming the growth rate found in James Foreman-Peck and Pedro Lains, 2000, �EuropeanEconomic Development: The Core and the Southern Periphery, 1870-1910,� in Şevket Pamukand Jeffrey G. Williamson, eds., The Mediterranean Response to Globalization Before 1950,Routledge, New York.
Data for Peru after 1900 come from Maddison op. cit. Before this date it is assumed thatPeru grew at the same year-on-year rate as did our estimates of Argentine GDP per capita.
Data for the Philippines after 1900 come from Maddison op. cit. Before this date it isassumed that Philippine GDP per capita grew at the same year-on-year rate as our estimates forSiam.
25
Estimates for Serbia after 1890 come from Foreman-Peck and Lains, op. cit. Before 1890GDP per capita is assumed to grow at the same year-on-year rate as it did between 1890 and1913.
Estimates for Turkey after 1913 come from Maddison. Before this date it is assumed thatGDP per capita grew at the same year-on-year rate as did estimates of Turkish real wages fromJeffrey Williamson, 2000, �Real wages and relative factor prices around the Mediterranean,1500-1940,� in Şevket Pamuk and Jeffrey G. Williamson, eds. The Mediterranean Response toGlobalization Before 1950, Routledge, New York.
Data for Uruguay after 1882 comes from Maddison op. cit. Before this date it is assumed thatUruguay grew at the same year-on-year rate as did our estimates of Argentine GDP per capita.GDP for Uruguay is taken from Mitchell (1993) for the period 1935-1940. Annual GDP percapita estimates 1914-1934 are calculated by assuming that Uruguay deviated from her GDP percapita trend (between the benchmark years of 1914, found in Clemens and Williamson (2000),and 1935, found in Mitchell) in the same way that Argentina did.
Data for a small remaining number of missing years are geometrically interpolated.
Terms of Trade Index (or the Net Barter Terms of Trade - NBTT)Existing data series were employed for the terms of trade for the US, the UK, France,
Germany, Sweden, Italy and Austria. Terms of trade for Austria-Hungary after 1882 are found inScott M. Eddie, 1977, �The Terms and Patterns of Hungarian Foreign Trade, 1882-1913,�Journal of Economic History, 37(2)(June):329-358. An index for 1876-1882 is constructed fromindices of the physical quanta and values of exports and imports given in Statistik desAuswärtigen Handels des Österreichisch-Ungarischen Zollgebiets im Jahre 1891, StatistischenDepartement im K. K. Handelsministerium, Vienna, 1893, pp. LXVIII-LXIX. For the period1865-1875 the same source reports only export and import values, not physical quanta. Since thequanta display extremely stable trends during 1876-1892 (unlike the values, which are subject tothe vagaries of prices), the quanta for 1865-1875 are extrapolated assuming the same, stablegrowth rate observed on 1876-1892. Combining these estimates with the trade value figuresgiven for 1865-1875 yield a ToT estimate for this period. Terms of Trade for France 1870-1896come from Charles Kindleberger, 1956, The Terms of Trade: A European Case Study, MITTechnology Press, Cambridge, Table 2-1, pp. 12-13. This is linked to a series from 1896-1913found in P. Villa, 1993, Une Analyse Macroéconomique de la France au XXeme Siècle, CNRSEditions, Monographies d�Économetrie, Paris, pp. 445-6. German ToT for the entire period comefrom Walther G. Hoffmann, 1965, Wachstum der Deutschen Wirtschaft seit der Mitte des 19Jahrhunderts, Springer-Verlag, Berlin, Table 134, col. 1, p. 548. Italy�s terms of trade with GreatBritain are taken as a proxy for overall Italian terms of trade. The former are found in I. A.Glazier, V. N. Bandera, and R. B. Berner, 1975, �Terms of Trade between Italy and the UnitedKingdom 1815-1913,� Journal of European Economic History, 4(1)(Spring): 5-48. Sweden�sterms of trade are taken from Simon Kuznets, 1996 [originally published 1967], �QuantitativeAspects of the Economic Growth of Nations: X. Level and Structure of Foreign Trade: Long-Term Trends,� reprinted in C. Knick Harley, ed. The integration of the world economy, 1850-1914, Volume 1, Elgar Reference Collection: Growth of the World Economy Series, Vol. 3.Cheltenham, U.K, Table 12, p. 150. United States ToT are from Jeffrey G. Williamson, 1964,American Growth and the Balance of Payments 1820-1913, University of North Carolina Press,Chapel Hill, North Carolina, Table B4, p. 262.
For the remaining countries, a NBTT series was calculated from original sources. Note thatthe NBTT is simply the ratio of export prices to import prices, each weighted appropriately.Mathematically,
26
∑∑
⋅
⋅= M
iMit
Xij
Xijt
jt wp
wpNBTT
for product i, country j, and period t. Note that in this formulation the numerator, the exportprice index, is country-specific while the denominator, the import price index, is not. This is asimplification employed in this paper due to (i) the limited quality and quantity of data on importsand import prices to countries in the periphery, and (ii) the similarity observed, in what recordsare available, between the composition of imports to developing countries. While detailed dataon exports weights and prices are available for virtually all of the countries and all of the years inour sample, import data are much more limited. These limitations and their consequences arediscussed below.
Export Weights. For the purposes of this study, export weights have been calculated byindividual country using the current value of major commodity exports and fixed weights. Theuse of a fixed set of weights is essential for disentangling price from quantity movements. Ofcourse, any such approach is fundamentally flawed, not least because over a long period of timethe mix of major commodity exports can shift significantly. A compromise position was taken bychanging the export weights at approximately 20-year sub-periods. These sub-periods are 1870-1890, 1890-1913, 1913-1929, and 1930-1950, and within these the weights are calculated usingsample year data. Export values for major commodities for Canada, Brazil, Chile, Colombia,Cuba, Mexico, Paraguay, and Peru are taken from Mitchell, International Historical Statistics TheAmericas 1750-1993, p.506 ff. Table E3. The same data for Australia, Burma, Ceylon, Egypt,India, Indonesia, Japan, Philippines, Siam, Turkey and New Zealand come from Mitchell,International Historical Statistics Africa, Asia and Oceania 1750-1993, p.637 ff. Table E3. Maincommodity exports for Denmark, Greece, Norway, Portugal, Spain and Sweden were calculatedfrom Statistical Abstract for Principal and Other Foreign Countries, London 1876-1912 and DieWirtschaft des Auslandes, Statistisches Reichsamt Berlin 1928.
Export Prices. Export prices are quoted in foreign markets (wherever possible, in the UK),rather than domestic ones. Wholesale Prices for Wheat, Maize, Rice, Beef, Butter, Sugar, Coffee,Tea, Iron, Copper, Tin, Lead, Coal, Cotton, Flax, Hemp, Jute, Wool, Silk, Hides, Nitrate, PalmOil, Olive Oil, Linseed, Petroleum, Indigo and Timber are taken from Sauerbeck, Prices ofCommodities and Precious Metals, Journal of the Statistical Society of London, vol. 49/3September 1886 Appendix C, for the years 1860-85. Sauerbeck, Prices of Commodities Duringthe Last Seven Years, Journal of the Royal Statistical Society, vol.56/2 June 1893 p.241 ff., for theyears 1885-1892, Sauerbeck, Prices of Commodities in 1908, Journal of the Royal StatisticalSociety 72/1 Mar 1909 for the years 1893-1908. Sauerbeck, Wholesale Prices of Commodities in1929, Journal of the Royal Statistical Society, vol. 93/2 p. 282 ff, 1930 for the years 1908-1929.Sauerbeck, Wholesale Prices of Commodities in 1916, Journal of the Royal Statistical Society,vol. 80/2 p. 289 ff. for the years 1908- 1916. Sauerbeck, Wholesale Prices in 1950, Journal of theRoyal Statistical Society, vol. 114/3 1951, p. 417 ff. for the years 1916-50. Prices for Cocoa,Crude Oil, Rubber, Tobacco and Zinc are taken from Dodd, Historical Statistics of the UnitedStates from 1790-1970, University of Alabama Press. Prices for Fruits and Nuts 1880-1914 aretaken from Critz, Olmsted and Rhode, International competition and the development of the driedfruit industry 1880-1930 table 8.2, in Pamuk and Williamson, 2000. Prices for Opium 1860-1906Ahmad Seyf, Commercialization of Agriculture: Production and Trade of Opium in Persia,1850-1906 Table 4, International Journal of Middle East Studies, 1984. Prices for Beans & BeanProducts were calculated from Liang-Lin, China�s Foreign Trade Statistics 1864-1949, HarvardUniversity Press 1974, p.80 ff.
Import Weights. A single set of import weights is employed for all countries in the sample.Import data, unlike that of exports, is almost uniformly poor, in particular in countries outside theEuropean Core. Traditionally, studies of country terms of trade have compensated for this lack of
27
data through the use of British export data as a proxy for the imports of less developed nations. This approach is undesirable given that the composition of British exports can hardly beconsidered representative of the imports of developing countries as a whole, and because the useof current-year weights means that movements reflect changes in composition, not just prices. Asan alternative, however, we employ a fixed index of non-primary goods from US statistics. Thisimport index, like the British one, is country invariant. In the end, the differences are notmaterial; the two series are almost identical (probably due to the heavy content of metals andtextiles in both indices). This US manufactured export statistic is a weighted sum of the prices oftextiles (55%), metals (15%), machinery (15%), building materials (7.5%), and chemicals andpharmaceuticals (7.5%). Obviously a fixed weighting for all developing nations isunrepresentative of their particular import mix (but while not representative of the specific importmix of the country, such a metric may be relevant for measuring the changing value of thecountry's exports versus a fixed package of manufactured products available for import. In thissense our terms of trade represent the purchasing power of local commodities in terms of rich-country goods.) Moreover, a review of each nation's external commerce documents turns upremarkably similar import compositions. For the years 1870-1900, import composition forAustralia, Canada, Ceylon, India and New Zealand was examined from Statistical abstract for theseveral colonies and other possessions of the United Kingdom no.1-40, 1863-1902. Data forBurma comes from Terulo Saito and Lee Kin Kiong Statistics on the Burmese Economy,Singapore 1999 pp 177, table VII-4. Import weights for China, Denmark, Egypt, Greece, Japan,Norway, Portugal, and Russia were calculated from Statistical Abstract for Principal and OtherForeign Countries, London 1876-1912 no. 13. Data for the Philippines are taken from QuarterlySummary of Commerce of the Phillipine Islands, Washington 1908 p.27 for the year 1893. Importcomposition for Serbia before 1914 is recorded in Sundhaussen, Historische Statistik Serbiens1834-1914, Munich 1989 pp. 352-355. Main imports for Turkey are calculated from Mulhall,Dictionary of Statistics, London 1892 p. 145, for the year 1888. For the years 1900-1940, importweights for Australia, Canada, Ceylon, India, New Zealand are calculated for several referenceyears from Statistical abstract for the several British self-governing dominions, colonies,possessions, and protectorates no.41-53, 1903-1915, Statistical abstract for the several Britishoversea dominions and protectorates no.54-59, 1917-1927, Statistical abstract for the BritishEmpire no.60-68, 1929-1938, Statistical abstract for the British Commonwealth no.69-70, 1945-1947 and Statistical abstract for the Commonwealth (trade statistics) no.71-72, 1948-1951.Composition of main imports for reference years after 1900 for Argentina, Chile, Greece,Indonesia, Japan, Mexico, Norway, Portugal, Russia, Serbia, Spain, Siam, Uruguay comes fromDie Wirtschaft des Auslandes 1900-1927, Berlin 1928. Data for Burma comes from Terulo Saitoand Lee Kin Kiong Statistics on the Burmese Economy, Singapore 1999 pp 177, table VII-4.Data for the Philippines is taken from Foreign Commerce of the Phillipine Islands,Washington1912-1913 for the reference years 1907, 1908 and 1910. Composition of main imports for Turkeywas calculated from Annuaire Statistique , Republique Turque, vol.1 pp. 103, 106 vol. 3 pp. 313,314 for the years 1923, 1926 and 1929.
Import Prices: US price series for textiles, metals, machinery, building materials, andchemicals and pharmaceuticals come from Dodd, Historical Statistics of the United States from1790-1970, University of Alabama Press.
An Additional Note on Import and Export Price Data. UK and US prices are employed inthe theory that the prices in these large, integrated and (in the UK, at least) unprotected marketswould supply us with a relatively reliable "world" price index for each commodity group. Achief disadvantage of using such world price indices, however, is that home market prices in eachcountry may diverge from the world ones in the short and even long term. This may be becauseof differences in product features and quality, because of variations in the composition of theproducts within a category, or because of less-than-perfect market integration combined withlocal market conditions and shocks. Kindleberger (1958) illustrates the wide divergence in the
28
prices of bulky products such as coal and lumber between two markets as closely integrated as theUS and UK. Another disadvantage of not using the home market price is the distortion created bychanges in transport costs. One would prefer a terms of trade measure that is independent oftransport costs. In a moment we will discuss the adjustments made to our terms of trade figuresto account for transport cost changes. Such adjustments as we can make, however, cannot trulyrepresent actual freight-adjusted prices. Overall, though, we feel the advantages of employingworld price indices outweigh these disadvantages. First and foremost, home market prices are nottypically on hand for the periods and countries in question. Rather, only the somewhat lessdesirable unit prices (calculated as the value of imports divided by the volume) are available. Second and more important, we believe UK and US market prices to be more reliable, accurateand comparable given the quality of reporting (at the time) and the quality of scholarship on theseprices since then. Third, to the extent that commodity markets are well integrated worldwide, theUK and US market prices should approximate the world price. This is especially true because weare interested in price changes, not levels. To the extent that UK and US prices move in similardirections and similar magnitudes to prices in the rest of the world, these "world" price indiceswill more or less represent price changes relative to an index year in other nations. We believethis to be a reasonable and necessary assumption. Fourth, these foreign market price indiceswould have been available to (and probably used) by industrialists and policymakers throughoutthe period in question. Accordingly, for questions of policy response (and perhaps price setting)foreign market indices may be a more appropriate data source than home market ones. Fifth, theuse of a world price index harmonizes and simplifies construction of the indices, enabling us toexamine a wider sample of countries at the cost, perhaps, of precision. Fifth, by measuring boththe export and import price indices in a common currency, we eliminate any inflationary biasfrom the figures.
29
Appendix 2: Additional Regression Results
Table A2: Regression Results without X/GDP, 1870-1938
(1) (2)Core Periphery
ln GDP per capita -3.125 -2.984[1.238]** [1.294]**
% ToT 0.331 3.392[0.320] [1.505]**
% ToT x PP/X -0.003 -0.035[0.004] [0.015]**
sd (ToT) 0.011 -0.053[0.033] [0.038]
PP/X -0.032 -0.106[0.023] [0.045]**
% Kids in School -0.019 -0.048[0.029] [0.054]
Constant 29.215 32.453[10.184]*** [10.386]***
Observations 110 96R-squared 0.4 0.43
Robust standard errors in brackets* significant at 10%; ** significant at 5%; *** significant at 1%
Figure 1: Trends in the Terms of Trade by Region
60
80
100
120
140
160
1870
1872
1874
1876
1878
1880
1882
1884
1886
1888
1890
1892
1894
1896
1898
1900
1902
1904
1906
1908
1910
1912
1914
1916
1918
1920
1922
1924
1926
1928
1930
1932
1934
1936
1938
1940
Term
s of
Tra
de In
dex
(190
0=10
0 )
European "Frontier" Offshoots Asia & the Middle East Latin America European Industrial Latecomers Industrial Leaders
Figure 2: Terms of Trade Volatility by Country
0
5
10
15
20
25
30
35
40
45
50
Swed
en UK
Fran
ce
Rus
sia
Den
mar
k
Indi
a
Can
ada
Arge
ntin
a
New
Zea
land
Spai
n
Uru
guay
USA
Burm
a
Peru
Serb
ia
Japa
n
Nor
way
Chi
le
Aust
ria-H
unga
ry
Turk
ey
Aust
ralia
Ger
man
y
Italy
Siam
Phili
ppin
es
Mex
ico
Portu
gal
Indo
nesi
a
Egyp
t
Chi
na
Gre
ece
Cub
a
Cey
lon
Braz
il
Col
ombi
a
Std
Dev
of T
erm
s of
Tra
de In
dex
1870-1889 1890-1913 1920-1939
Figure 3: Terms of Trade Volatility by Primary Product Commodity
0
20
40
60
80
100
120
140C
ork
Coc
oa
Cop
per
Pal
m O
il
Rub
ber
Mea
t
Lead
Lins
eed
But
ter
Tea
Hid
es
Zinc Iron
Whe
at
Silk Tin
Silv
er
Cof
fee
Nitr
ate
Frui
ts/N
uts
Hem
p
Live
stoc
k
Win
e/S
pirit
s
Jute
Tim
ber
Woo
l (Is
lay)
Mai
ze
Cot
ton
Pet
rol
Woo
l (M
erin
o)
Ric
e
Sug
ar (W
est I
ndie
s)
Sug
ar (J
ava)
Oliv
e O
il
Toba
cco
Flax
Fish
Bea
n P
rodu
cts
Std
Dev
of C
omm
odity
Pric
e In
dex,
187
0-19
39
Figure 4: GDP per capita Growth and ToT Volatility, 1870-1939
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
3C
UB
BU
R
IND
THA
CE
Y
CH
N
AU
S
SP
A
PO
R
PH
L
CO
L
IDN
SR
B
BR
A
UK
ME
X
GR
C
NZD
RU
S
ITA
FRA
US
A
CA
N
AR
G
UR
U
TKY
GE
R
EG
P
DE
N
SW
E
CH
L
NO
R
JPN
AH
G
PE
R
Avg
Ann
ual G
row
th R
ate,
187
0-19
39
-10
-5
0
5
10
15
20
25
Std
Dev
Avg Annual Growth Rate Std Dev of ToT Index
Table 1: Export ConcentrationChief Export Commodities (as % of Total Exports)
Primary % Secondary % Top 2 Primary % Secondary % Top 2 Primary % Secondary % Top 2 Primary % Secondary % Top 2
European Industrial LatecomersAustria-Hungary . . . . . . . . . .Denmark . . . . . Butter 55% Meat 33% 88% Butter 46% Meat 45% 91% . . . . .Greece . . . . . Fruits/Nuts 59% Lead 14% 73% Tobacco 65% Fruits/Nuts 21% 86% . . . . .Italy . . . . . . . . . . . . . . .Norway Fish 52% Wood 48% 100% Fish 50% Wood 44% 94% Wood 61% Fish 39% 100% . . . . .Portugal Wine 62% Cork 13% 75% Wine 56% Cork 16% 73% Wine 54% Fish 30% 85% . . . . .Serbia Livestock 55% Fruits/Nuts 18% 73% Livestock 49% Grain 29% 77% Wood 35% Livestock 29% 64% . . . . .Spain Wine 53% Fruits/Nuts 12% 64% Iron 24% Fruits/Nuts 22% 46% . . . . .Russia Grain 65% Flax 14% 79% Grain 63% Wool 11% 73% Grain 28% Petroleum 20% 48% Wood 39% Grain 24% 63%
Average 57% 21% 78% 51% 24% 75% 48% 31% 79% 39% 24% 63%
European "Frontier" OffshootsArgentina Wool 56% Hides 31% 87% Wool 35% Wheat 23% 58% Wheat 31% Maize 20% 51% Maize 25% Meat 22% 47%Australia Wool 89% Wheat 9% 98% Wool 73% Meat 11% 84% Wool 59% Wheat 25% 85% Wool 59% Wheat 17% 76%Canada Wood 54% Wheat 42% 96% Wood 36% Wheat 32% 68% Wheat 49% Wood 17% 66% Wheat 32% Paper 19% 51%New Zealand Wool 99% Butter 1% 100% Wool 59% Meat 31% 89% Wool 36% Meat 34% 70% Meat 36% Wool 34% 69%Uruguay Hides 44% Wool 30% 74% Wool 40% Hides 32% 72% Meat 41% Wool 39% 80% Wool 54% Meat 31% 85%
Average 68% 23% 91% 49% 26% 74% 43% 27% 70% 41% 25% 66%
Latin AmericaBrazil Coffee 70% Sugar 16% 86% Coffee 65% Rubber 26% 91% Coffee 83% Sugar 6% 88% Coffee 68% Cotton 24% 92%Chile Copper 68% Nitrate 32% 100% Nitrate 81% Copper 19% 100% Nitrate 75% Copper 25% 100% Copper 62% Nitrate 38% 100%Colombia Tobacco 61% Coffee 39% 100% Coffee 92% Tobacco 8% 100% Coffee 98% Tobacco 2% 100% Coffee 74% Petroleum 26% 100%Cuba . . . . . Sugar 70% Tobacco 30% 100% Sugar 90% Tobacco 10% 100% Sugar 87% Tobacco 13% 100%Mexico Silver 92% Coffee 7% 99% Silver 75% Copper 11% 86% Petroleum 69% Silver 16% 85% Silver 31% Petroleum 31% 62%Peru Sugar 48% Silver 26% 74% Sugar 32% Silver 23% 55% Sugar 31% Cotton 28% 59% Petroleum 40% Cotton 27% 67%
Average 68% 23% 91% 59% 23% 81% 59% 21% 80% 51% 26% 76%
Asia & the Middle EastBurma Rice 80% Wood 20% 100% Rice 83% Wood 10% 93% Rice 70% Petroleum 18% 87% Rice 48% Petroleum 36% 83%Ceylon Coffee 99% Tea 1% 100% Tea 98% Coffee 2% 100% Tea 67% Rubber 33% 100% Tea 75% Rubber 25% 100%China Tea 44% Raw Silk 29% 73% Raw Silk 45% Tea 21% 66% Raw Silk 57% Textiles 13% 69% Raw Silk 21% Eggs 21% 43%Egypt Cotton 100% . . 100% Cotton 100% . . 100% Cotton 100% . . 100% Cotton 100% . . 100%India Opium 28% Cotton 27% 55% Textiles 30% Rice 21% 51% Cotton 33% Textiles 30% 63% Cotton 32% Textiles 28% 59%Indonesia . . . . . Sugar 41% Coffee 19% 60% Sugar 44% Petroleum 21% 65% Rubber 29% Petroleum 25% 54%Japan Raw Silk 100% . . 100% Raw Silk 55% Textiles 44% 99% Raw Silk 52% Textiles 47% 99% Textiles 55% Raw Silk 34% 89%Philippines Sugar 47% Hemp 35% 81% Hemp 73% Sugar 17% 90% Sugar 47% Hemp 33% 80% Sugar 65% Hemp 16% 81%Siam Rice 100% . . 100% Rice 100% 100% Rice 84% Tin 15% 99% Rice 66% Tin 21% 87%Turkey Fruits/Nuts 33% Wool 18% 50% Fruits/Nuts 36% Raw Silk 23% 59% Tobacco 45% Fruits/Nuts 30% 75% Fruits/Nuts 37% Tobacco 35% 72%
Average 70% 22% 84% 66% 20% 82% 60% 26% 84% 53% 27% 77%
Total Average 68% 22% 89% 61% 22% 82% 59% 23% 81% 52% 26% 77%
Sources: See data appendix under Terms of Trade -- Export Weights
1878-1882 1898-1902 1920-1924 1934-1938
Table 2: Exports by CategoryChief Export Classifications (as % of Total Exports)
Ag/Min Fuels Mftrs Ag/Min Fuels Mftrs Ag/Min Fuels Mftrs Ag/Min Fuels Mftrs
European Industrial LatecomersAustria-Hungary . . . . . . . . . . . .Denmark 100% 0% 0% 100% 0% 0% 100% 0% 0% . . .Greece . . . 100% 0% 0% 100% 0% 0% . . .Italy . . . . . . . . . . . .Norway 100% 0% 0% 100% 0% 0% 100% 0% 0% . . .Portugal 100% 0% 0% 91% 0% 9% 100% 0% 0% . . .Serbia 100% 0% 0% 100% 0% 0% 100% 0% 0% . . .Spain 100% 0% 0% 100% 0% 0% . . . . . .Russia 100% 0% 0% 90% 10% 0% 68% 20% 12% 71% 12% 17%
Average 100% 0% 0% 97% 1% 1% 95% 3% 2% 71% 12% 17%
European "Frontier" OffshootsArgentina 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%Australia 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%Canada 100% 0% 0% 97% 0% 3% 95% 0% 5% 94% 0% 6%New Zealand 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%Uruguay 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%
Average 100% 0% 0% 99% 0% 1% 99% 0% 1% 99% 0% 1%
Latin AmericaBrazil 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%Chile 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%Colombia 100% 0% 0% 100% 0% 0% 100% 0% 0% 74% 26% 0%Cuba 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%Mexico 100% 0% 0% 100% 0% 0% 31% 69% 0% 69% 31% 0%Peru 100% 0% 0% 100% 0% 0% 81% 19% 0% 60% 40% 0%
Average 100% 0% 0% 100% 0% 0% 85% 15% 0% 84% 16% 0%
Asia & the Middle EastBurma 100% 0% 0% 94% 6% 0% 82% 18% 0% 64% 36% 0%Ceylon 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%China 94% 0% 6% 91% 0% 9% 87% 0% 13% 90% 0% 10%Egypt 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%India 90% 0% 10% 75% 0% 25% 70% 0% 30% 72% 0% 28%Indonesia . . . 97% 0% 3% 79% 0% 21% 75% 0% 25%Japan 100% 0% 0% 56% 0% 44% 53% 0% 47% 36% 0% 64%Philippines 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%Siam 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%Turkey 100% 0% 0% 100% 0% 0% 100% 0% 0% 100% 0% 0%
Average 98% 0% 2% 91% 1% 8% 87% 2% 11% 84% 4% 13%
Total Average 100% 0% 0% 97% 1% 2% 92% 5% 4% 84% 8% 8%
Note: Ag/Min=Agricultural and mineral products, Fuels=Oil, gas and coal, and Mftrs=Manufactured products (primarily textiles and metal pro MftrsSources: See data appendix under Terms of Trade -- Export Weights
1878-1882 1898-1902 1920-1924 1934-1938
Table 3: Was the Terms of Trade Correlated with Economic Growth 1870-1938?
(1) (2) (3) (4)
Core Periphery Core Peripheryln GDP per capita -3.119 -2.711 -5.971 -2.642
[1.252]** [1.315]** [2.271]** [1.404]*% ToT 0.325 2.650 0.502 2.873
[0.356] [1.351]* [0.475] [1.416]**% ToT x PP/X -0.003 -0.029 -0.006 -0.036
[0.004] [0.014]** [0.005] [0.015]**% ToT x X/GDP 0.001 0.017 0.001 0.045
[0.009] [0.006]** [0.012] [0.015]***sd (ToT) 0.011 -0.071 0.026 -0.160
[0.033] [0.041]* [0.047] [0.059]**sd (ToT) x X/GDP -0.000 0.002 -0.000 0.009
[0.003] [0.002] [0.004] [0.004]**% Kids in School -0.019 -0.029 -0.025 -0.106
[0.031] [0.053] [0.042] [0.060]*PP/X -0.032 -0.072 -0.041 -0.128
[0.024] [0.046] [0.042] [0.053]**Constant 29.162 27.331 54.948 33.328
[10.441]*** [10.193]*** [20.592]** [10.830]***Observations 110 96 72 64R-squared 0.40 0.48 0.48 0.60Dependent variable is decadal GDP per capita growth.Notes: Robust standard errors in brackets. Excludes war years. Core = France, Germany, Sweden, UK,US; Australia, Canada, New Zealand; Austria-Hungary, Denmark, Greece, Italy, Norway, Portugal, Serbia, Spain, Russia; Argentina, Uruguay. Periphery = Brazil, Colombia, Chile, Cuba, Mexico, Peru; Burma, Ceylon, China, Egypt, India, Indonesia, Japan, the Philippines, Siam, Turkey. All estimatesare with country and time fixed effects.* significant at 10%; ** significant at 5%; *** significant at 1%
1870-1938 1890-1938
Table 4: The Impact of the Terms of Trade 1870-1938
(1) (2) (3) (4)
Core Periphery Core PeripheryMean Values
% GDP 1.39 0.82 1.56 0.67% ToT 0.10 -0.86 -0.20 -1.58% ToT x PP/X 2.42 -79.01 -27.83 -149.55% ToT x X/GDP -0.79 -13.86 -5.13 -23.48sd (ToT) 8.52 11.42 9.18 10.84sd (ToT) x X/GDP 97.63 154.97 114.37 156.45PP/X 77.57 93.55 76.13 92.91X/GDP 12.20 12.67 13.55 12.86
Coefficient Estimates (Table 3)% ToT 0.325 2.650 0.502 2.873% ToT x PP/X -0.003 -0.029 -0.006 -0.036% ToT x X/GDP 0.001 0.017 0.001 0.045sd (ToT) 0.011 -0.071 0.026 -0.160sd (ToT) x X/GDP -0.000 0.002 -0.000 0.009
Marginal Impact% ToT 0.104 0.153 0.059 0.107sd (ToT) 0.011 -0.046 0.026 -0.044
Actual Impact% ToT 0.024 -0.212 0.059 -0.211sd (ToT) 0.094 -0.501 0.239 -0.326% and sd combined 0.118 -0.713 0.298 -0.538
Marginal Impact : adds up marginal effect of terms of trade using PP/X and X/GDP means as weights for the interacted terms Actual Impact : adds up actual effect of terms of trade using %ToT and sd ToT means, as well as the means of the interacted terms, that applied to the region in question.
Based on Table 31870-1938 1890-1938